Figure 3
From: Predictive scale-bridging simulations through active learning
Top row: We show the performance of the ML surrogate model for all the mutual diffusion points in the test dataset, which were generated by randomly sampling 1100 points from a 5-D input parameter space. From left to right: Deuterium–deuterium, Deuterium–Argon, and Argon–Argon mutual diffusions. The points selected for the MD verification are circled in red. Bottom row: Locations of MD calls during an interfacial mixing simulation. White = MD, Blue = surrogate model, Black = database duplicate. The simulation on the left used a uniform temperature profile, while the case on the right is artificially heated to push the state of the system away from the initial training data used for the surrogate model32.
